📈 economics

Reputation and Disclosure in Dynamic Networks

This paper analyzes a disclosure protocol in dynamic networks where public review dates and retention rules act as censoring events, demonstrating how separating custody from audit information and utilizing Ornstein-Uhlenbeck benchmarks enables finite-time resolution of reputational uncertainty and reveals network values created by common-record censoring that pairwise formation strategies overlook.

Original authors: I. Sebastian Buhai

Published 2026-05-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: I. Sebastian Buhai

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Sealed Envelope" Game

Imagine a high-stakes game involving a sealed envelope containing a secret document (like a compliance report or a piece of hard evidence). This envelope is passed from person to person before it finally reaches a Decision Maker (like a judge, a CEO, or a regulator).

The twist is that the envelope is docketed. This means everyone knows:

The envelope exists.
Who currently holds it.
When the next scheduled "inspection" (review date) will happen.

However, no one knows what is inside the envelope until the very last person decides to open it.

The Core Problem: Silence is Loud

In this game, the people holding the envelope (the intermediaries) have a choice at every inspection date:

Pass it on: They give the envelope to the next person in line.
Keep it (Retain): They hold onto it for another round.
Open it: If they are the last person, they reveal the contents to the Decision Maker.

The Paper's Insight: In the past, if someone stayed silent, you didn't know why. Maybe they hadn't looked yet, or maybe they were hiding something. But in this paper's system, because the inspection dates are public, silence becomes a signal.

If a person keeps the envelope at a scheduled inspection, they are effectively saying: "I am keeping this because the secret inside is in a specific range that I don't want to reveal yet." By keeping it, they are censoring (hiding) the bad news, but the act of hiding it tells the public exactly what kind of news it is (e.g., "It's not terrible, but it's not great either").

The "Gaussian" Analogy: The Foggy Window

The paper uses complex math (Gaussian distributions) to describe how people guess what's inside.

Imagine the secret is a number, and everyone has a "foggy window" through which they try to guess that number.

The Public View: The public has a general idea of the number based on a "public signal" (like a weather report).
The Holder's View: The person holding the envelope has a clearer view (a private signal).

When the holder decides to keep the envelope, they are telling the public: "The number is in this specific slice of the fog."

If they pass the envelope, they are saying: "The number is NOT in that slice; it's somewhere else."

The paper proves that if the holders use simple rules (like "I'll keep it if the number is between 10 and 20"), the public can calculate the exact probability of what's inside, even without seeing it. The "fog" gets thinner with every decision, but the shape of the possible numbers changes in a predictable way.

The "Network Race": Two Paths, One Secret

The paper then introduces a more complex scenario: The Race.

Imagine the sender has two different routes to get the envelope to the Decision Maker:

Route A: Sender $\to$ Person A $\to$ Decision Maker.
Route B: Sender $\to$ Person B $\to$ Decision Maker.

The sender puts copies of the same sealed envelope on both routes.

The Magic of the Race:
If Person A decides to keep the envelope at their inspection, it doesn't just affect Route A. Because the envelopes are identical, Person B (on Route B) instantly learns that the secret cannot be in the range Person A is hiding.

This is called "Common-Record Censoring."

Analogy: Imagine two runners in a race. If Runner A stops to tie their shoe (retains the file), Runner B doesn't just know Runner A stopped; Runner B instantly knows that Runner A's shoe-tying range is off-limits for the secret. This changes the strategy for Runner B immediately, even though Runner B hasn't looked at their own envelope yet.

The Main Findings

Delay is Informative: In this system, waiting isn't just "doing nothing." Every time someone keeps the file, they are narrowing down the possibilities for everyone else. The "delay" itself is a message.
Old Files Must Be Released: The paper shows that if a file sits around for too long, the person holding it will eventually be forced to release it. Why? Because holding onto an old file hurts their reputation (people start thinking they are hiding something bad), and the benefit of waiting gets smaller as the file gets older.
Networks Can Be "Too Sparse": This is the most surprising part.
- The Problem: Two people might decide not to build a second route (like Route B above) because, individually, it doesn't seem worth the cost. They only see the direct benefit of having a second path.
- The Missed Value: They fail to see the Network Value. By having two paths, the "censoring" effect happens faster. If Person A holds, Person B learns instantly. This speeds up the whole system and helps the Decision Maker get the truth sooner.
- The Result: Private people might agree to a network that is too simple (only one path) because they can't price the "information speed" that comes from having two paths competing. A planner (a boss) would want the second path, but the individuals won't build it.

Summary in One Sentence

This paper shows that when people hold onto a secret document in a public system, their silence reveals exactly what they are hiding; furthermore, having two competing paths for the same secret makes the whole system smarter and faster, but private individuals often fail to build that second path because they don't realize how much value the "competition" creates for everyone else.

Technical Summary: Reputation and Disclosure in Dynamic Networks

Problem and Motivation
The paper investigates the informational dynamics of hard evidence disclosure within institutions characterized by public docketing. In many compliance, audit, or legal settings, a sealed record is known to exist, its custody is public, and review dates are scheduled and observable, yet the content remains hidden until a terminal disclosure occurs. The central problem is how "delay" functions as a signal in such environments. Unlike standard disclosure models where silence is ambiguous, the paper posits that in a public docket system, retention at a scheduled review date is an informative censoring event. It reveals that the hidden state lies within the holder's retention region, thereby ruling out states where the holder would have relayed or disclosed the record. The paper seeks to formalize this censoring mechanism, characterize the resulting equilibrium strategies, and analyze the network implications when multiple certified routes carry the same record.

Methodology and Model Framework
The author constructs a continuous-time dynamic game involving a decision maker (DM), experts, and relay intermediaries.

Fundamentals: The underlying state $X_t$ follows an Itô diffusion, specialized in the main benchmark to an Ornstein-Uhlenbeck (OU) process. A public signal $Y_t$ provides noisy information about $X_t$ .
Information Structure: A sealed record $\varrho_\tau = (X_\tau, \tau)$ is docketed at time $\tau$ . The current custodian privately observes the content, forming a posterior mean $\mu^i_t$ , while the public maintains a posterior mean $m_t$ . The hidden object is the scalar gap $d^i_t = \mu^i_t - m_t$ .
Protocol: The system features public custody logs, scheduled review dates, and distinct audit marks. At a review date, a custodian chooses to retain, relay, or disclose. Retention censors the support of the hidden gap; relay or disclosure reveals the complementary event.
Belief Updating: The core technical innovation is the derivation of an exact filtering recursion. The public posterior is not a simple mixture over latent histories but a Gaussian likelihood kernel conditioned on a feasible support set $K_t$ . This support set evolves deterministically via affine transport between reviews and undergoes truncation or "hole-creation" at review dates.
Equilibrium Concept: The analysis focuses on pure Markov Perfect Bayesian Equilibria (MPBE) using interval policies. Strategies are defined by retention intervals within the feasible support. The author employs a verification theorem using the generator of the joint state process to confirm equilibrium conditions.

Key Contributions

Exact Censoring Recursion: The paper provides a rigorous derivation of how public docketing transforms delay into a precise support restriction. It demonstrates that with interval strategies, the feasible support of the hidden gap remains a finite union of intervals, allowing for an exact, finite-dimensional representation of the public state. This contrasts with models lacking public custody logs, where delay results in a complex mixture over latent review histories.
Compact Equilibrium Verification: The author establishes a benchmark verification theorem for the single-file path. By assuming an OU-quadratic structure and compact invariant domains, the paper proves the existence of MPBEs where retention strategies are defined by intervals. This framework supports "discipline-assisted" resolution, ensuring that old records are disclosed in finite time under reputational pressure.
Common-Record Censoring in Networks: The paper introduces a novel network externality: "common-record censoring." When two certified routes carry the same docketed record, a retention event on one route immediately restricts the feasible support of the hidden gap on the other route, even before the second route acts. This creates a network value that is distinct from simple racing or direct access.

Main Results

Filtering and State Representation: The exact public posterior is a Gaussian kernel conditioned on a support set $K_t$ . Under interval policies, $K_t$ is a finite union of intervals. The state space can be summarized by the public mean, the reference variance, the smoother block, reputation, custody, age, and the endpoints of the support intervals.
Resolution of Old Records: Under assumptions of review recurrence and reputational discipline (where skipping a review incurs a reputational cost), the paper proves that if a file remains unresolved past a deterministic age threshold, the next active review will induce forward movement (relay or disclosure) almost surely. This holds even with bias reversals among intermediaries.
Network Formation and Underconnection: In a two-route race, the paper demonstrates that the "common-record censoring" value creates a wedge between planner welfare and private incentives.
- The planner values the completion of a parallel branch because the censoring effect on one route improves the information structure for the other, raising the decision maker's expected payoff.
- Private agents, however, only internalize direct access, racing speed, and route-credit rents. They fail to internalize the cross-route censoring benefit.
- Consequently, the paper shows that pairwise stable networks can be strictly underconnected: a socially valuable parallel branch may fail to form because the private gains from the individual links (which are zero or negative in isolation) do not cover the costs, despite the joint branch generating positive social value through the censoring mechanism.

Significance and Claims
The paper claims to offer a precise characterization of how public institutional architecture (dockets, logs, scheduled reviews) alters the informational content of delay. It argues that separating custody from audit information allows for a tractable, exact Bayesian update where the support of the hidden state is tracked by finitely many endpoints.

The primary significance lies in the identification of a new network externality. The paper asserts that "common-record censoring" generates planner value that pairwise network formation mechanisms may miss. This suggests that stable networks can be inefficiently sparse not due to traditional coordination failures or externalities, but because the informational value of shared records across competing routes is not priced by individual link consent. The results imply that disclosure policy and network design are intrinsically linked through the public architecture of record-keeping, and that mechanisms relying solely on bilateral consent may fail to activate socially optimal disclosure channels.